Since April 2026, US officials have issued a stream of warnings about Chinese “distillation campaigns,” high-volume querying of US models to extract their capabilities, and have used those warnings to justify new restriction efforts aimed specifically at Chinese open-weight models. No nationwide ban on open-weight AI models exists in the US as of July 2026, but the policy fight over whether one should is real, active, and escalating.
What Restrictions Already Exist?
At the federal level, the “No DeepSeek on Government Devices Act” (H.R.1121, 119th Congress), introduced by Reps. Josh Gottheimer (D-NJ) and Darin LaHood (R-IL) along with 16 other co-sponsors, would bar DeepSeek from federal employee devices (Congress.gov, FedScoop). In the Senate, Sens. Bill Cassidy and Jacky Rosen introduced a bill requiring the Federal Acquisition Security Council to maintain a public list of AI models from “adversarial nations,” China, Russia, Iran, and North Korea, with agencies needing congressional or OMB exemption to use listed models (CyberScoop).
At the state level, Texas was first in February 2025, followed by New York on February 10, 2025, with Virginia, Iowa, South Dakota, and North Carolina also banning DeepSeek from government devices and networks (StateTech Magazine, NY Governor’s Office). Every restriction currently in force is concentrated in government devices, defense contractor systems, and intelligence-community networks. None of it amounts to a nationwide consumer or commercial ban.
Why a Full Ban Faces Real Limits
AI policy experts describe an outright ban on open-weight models as effectively impossible once the weights are already public and downloaded worldwide. There’s no mechanism to un-publish a file that millions of people already have a copy of. Legal experts have also flagged First Amendment concerns with prohibiting the distribution of publicly released model weights, meaning any broad ban would likely face a real constitutional challenge before it could take effect (TechTimes).
The Real Fight Is Narrower Than “Ban Open Models”
The live debate in Washington isn’t actually “should open-weight models exist.” It’s which specific models, generally tied to specific foreign, adversarial-nation origins, should be excluded from which specific systems: government devices, federal contractor networks, critical infrastructure. That’s a much narrower and more legally defensible target than banning the concept of open weights outright, and it’s the version of this fight that’s actually moving through Congress and state legislatures right now.
Frequently Asked Questions
Has the US banned open-weight AI models? No nationwide ban on open-weight AI models exists as of mid-2026. Restrictions so far are limited to specific contexts, government employee devices, defense contractor systems, and some state government networks, and target specific foreign-origin models like DeepSeek rather than open-weight models as a category.
Which US states have banned DeepSeek? Texas was first in February 2025, followed by New York on February 10, 2025, with Virginia, Iowa, South Dakota, and North Carolina also banning DeepSeek from government devices and networks.
What is the “No DeepSeek on Government Devices Act”? It’s a federal bill, H.R.1121 in the 119th Congress, introduced by Reps. Josh Gottheimer and Darin LaHood along with 16 other co-sponsors, that would bar DeepSeek from federal employee devices.
Could the US legally ban open-weight models entirely? It would face significant legal and practical hurdles. Policy experts note that once model weights are public and downloaded worldwide, a ban can’t actually remove them from circulation, and legal experts have raised First Amendment concerns about prohibiting distribution of already-published model weights.
What is a “distillation campaign” in this context? Since April 2026, US officials have used the term to describe high-volume querying of American AI models specifically to extract their capabilities and train competing models, and have cited it as justification for tighter restrictions on foreign open-weight models.
